Concepts and Glossary

Getting familiar with eva terms

Glossary

List of most common words in alphabetical order

TERM	DEFINITION
Admin	Profile with access to all eva resources
Agent Template	Pre-built and ready-to-use framework templates that can be adjusted to different use-cases.
API	Application programming interface, a communication protocol that helps different apps communicate with each other.
Automated tests	Process of making sure that a virtual agent answers an intent within expected parameters.
Bulk training	Massive import of intents to a virtual agent knowledge base or examples/utterances to an intent.
Channel	Platform where the virtual agent interacts with users.
Cell	Visual flow element that makes it easier to use the Workspace interface and to create and manage your virtual agent.
Cockpit	Name of the space that hosts all eva resources.
Cognitive engine	Software that processes human natural language.
CTA	Call to action. These are links or buttons on a page that encourage users to take a specific action. Sometimes it comes as an email form and a button with texts like “Subscribe”.
Dashboard	Feature in Cockpit where Admins and Editors can check the virtual agent’s performance through provided metrics.
Disambiguation	Resource used when building the knowledge base that removes possible ambiguities or requests help from the user to complement their interaction.
Dialog Manager	Main feature in eva, it's the workspace where you can create dialogs and build your virtual agent knowledge base.
Editor	Profile that can access all eva resources, except create, edit and delete users and virtual agents.
Entity	User interaction, usually associated with an adjective, noun, product, services, etc., that can modify or complement the Intent.
eva	NTT DATA virtual agent - Virtual agents building and managing platform.
eva NLP	NTT DATA proprietary Natural Language Processing (NLP) engine.
EVG	eva Voice Gateway
Evaluable Answer	Answer that can be evaluated by the user.
Handoff	Process when a human agent takes over the conversation, usually because the virtual agent doesn't understand the user, or the issue is not yet in the knowledge base.
Integration	eva allows integration with different cognitive engines, like eva NLP (NTT DATA), Watson (IBM), Dialogflow (Google), Luis (Microsoft), and others.
Intent	What the user has in mind when asking a question; represents its main purpose. Identifying these intents is critical to a virtual agent success for it’ll start the conversation.
Knowledge AI	Feature that interprets users’ questions and generate context-sensitive answers from documents uploaded to the platform.
Knowledge Base	The sum of all flows, intents, answers, services and trainings in a virtual agentt.
Metrics	Measure of a data to manage or analyze the virtual agent’s performance.
Not Expected	A fallback answer that works as a flow when the virtual agent doesn’t understand the user’s context.
Parameter	Value added to configure software behavior.
Property	A JSON code used in the technical text field used as commands in voice agents
Repository	The place where intents, answers, services, flows and trainings are stored.
Request	A request/report from the customer via eva support page (Jira).
Ticket	Number assigned to each request. It helps you track updates on the workflow.
Training	Name of the process that allows the virtual agent to learn to interpret user inputs; intent classification.
Transactional Answer	Answer that must be connected to an external API (depends on external sources). Your virtual agent will have to look elsewhere to answer your user and you have to show where.
User	The person who is talking to the virtual agent.
Utterance/Example	A sentence the user would say during a conversation with the virtual agent. It is the most important component of an intent.
Variants	Different ways user may ask for a subject that is part of a specific document in Automated Learning knowledge base.
Virtual Agent	A virtual agent capable of understanding human speech and to respond accordingly.
Webhook	A way of receiving information between two applications.
Welcome Message	The first message the virtual agent sends to start off chatbot-human interaction. It is an answer cell and works as a unique flow.
Workspace	Where you can visualize and design flows.

Main Concepts of eva

In a nutshell, everything you need to build a virtual agent in eva.

eva components diagrams

Dialog Manager

The Dialog Manager is the main feature, the canvas where you create the dialogs and build your virtual agent knowledge base.

These are the main characters in Dialog Manager:

1. Workspace

2. Flows

3. Cells

4. Training (if you are using eva NLP)

5. Knowledge AI

6. Dashboards - Analytics

Workspace

The Workplace is where you view and design the conversational flows. Here, you will add cells and string them together to give life to the dialog.

Dialog Cells

Conversational flows consist of cells, which are coordinated elements designated for specific functions within the flow. Learn more about each cell

Simulations of a flow building in the Workspace

Flows

When you connect cells, you create a conversational flow. Create flows with distinct use cases to create your virtual agent. Learn more about flows

Training

Training the virtual agent is important to improve its ability to understand what the user is asking and to be able to answer those questions accurately and efficiently. Training helps the agent learn language patterns, identify intentions and relevant entities, and provide appropriate responses.

The more the agent is trained, the better the results obtained.

More information about training using eva NLP:

• Training with eva NLP

• Using other NLP engines

Tests

Run the dialog to test it out on the dialog simulator. Click the chat widget icon in the bottom right corner. That's where you can simulate the dialog and test the flows.

Read more about how to test the flows

Simulate the conversation and test the flows

---

Dialog Cells

Intent cell

When designing a conversational flow you need to predict the user interactions and the virtual agent responses.

An intent cell will represent the users' interactions. It is the representation of a user's will, in other words, an action.

Usually it is associated with a verb. For example, when the user says he wants to buy something he is expressing a will, that is, an intention.

Since users can express the same will in different ways, the intention cell is composed of examples. The examples are all the possible ways the user can express the same will.

Examples of a purchase intention:

• I want to buy

• I want to acquire

• I want to make a purchase

Learn more about intents

Entity cell

Entities represents a user interaction, but unlike the Intent, it is usually associated with an adjective, noun, product, services, etc. In most cases, entities complement intentions.

For example, if the the user says: "I want to buy an Apple phone"

The terms "cell phone" and "Apple" will be Entities, because they represent a product and a brand.

Entities can also represent a very specific user interaction, such as an email. In cases where it is impossible to map all possible user interactions, such as an email list, we design a standardized term that will cater for this type of interaction. And for this we use standard type entity.

Learn more about entities

Answer cell

As the name suggests, the Answer cell will bring a virtual agent's feedback to a user's input.

In eva, you can also make your response evaluable. In the bottom corner, you can put a thumbs up and a thumbs down. This way you will know whether the user's feedback was positive or negative.

eva has six types of responses:

1) Simple Answers: the virtual agent predicts answers to user questions

b) Not Expected Answers: the answer given by the virtual agent to an unidentified user demand

c) FAQ: flows built based on answers to intentions

d) Dynamic Answers: through simple codes, brings personalized answers

e) Transactional Answers: when the answer will depend on an API

f) Channel-based Responses: responses that vary according to the channels

Learn more about answers

Jump cell

As the name suggests, the Jump function allows you to go from the last cell created to any other cell in any flow.

This is perfect to avoid repetition, as well as shortening and simplifying flows.

Learn more about jump cells

Service cells

Often, an answer to a user demand will require information that will necessarily be outside the eva, on other servers.

For example, informing a customer's balance, credit card account, closing a purchase on an ecommerce virtual agent, etc.

Here comes the Service cell, which connects the user to eva and the customer.

In it, there is a field to put Webhook. If you have never heard of this expression, it is a protocol used so that an application (APP) on the internet receives real-time information as soon as an event occurs in another web-based application.

Learn more about service cell

Wait-Input cell

"What is your birth date?", "write your name here", "put your account number here". The Input cell will be responsible for storing this information in the eva.

There are three types:

a) Date Template: allows the insertion of dates.

b) Time Template: allows the insertion of time (hour and minutes)

c) Customizable templates: you can enter locations, zip codes, among other data

Learn more about input cells

End cell

The End Cell works like an end point in a flow. By placing this feature, the virtual agent stops communicating at that point. That is, it will only resume the dialog if the user (the virtual agent's interlocutor) interacts again.

The End Cell is always used in a Jump flow, never in the User Journey (the main one).

Learn more about end cells

Code cell

The Code Cell allows you to perform some services without relying on APIs. It offers immense advantages in a flow.

Through the Code Cell it is possible to

• Manipulate objects

• Anticipate executions and actions

• Perform services without the need for APIs

• Reduce time and cost

Learn more about code cells

Rule cell

Imagine two users that have the same intention: checking their bank balance. However, one is already logged into the bank and the other is not. In cases like this, the Rule Cell comes in, to steer each user in the right direction.

This is a feature to make your dialog more assertive and much more precise, especially when Intentions and Entities can bring up multiple scenarios.

The language used is JavaScript.

Learn more about rule cells